Presentation
Learning Box Model using DSO.ai
DescriptionIn order to maximize design PPA, physical Implementation technology has advanced to handle every possible design knobs ranging through RTL, floorplan, SDC/UPF, library offering, power budget, etc. But overwhelming number of design knobs posed a new challenge for physical designers as reaching an optimal recipe within a given time-to-market is close to impossible for most modern size designs. In this paper, we propose a new ML based model, namely Learning-Box Model, to find optimal combination of application options (app-opt), specifically for P&R tools like FC/ICC2. Learning-Box Model minimizes conventional Cold-Start effort by using representative design combined with an optimal app-opt candidates. Simulation results showed proposed approach can the TNS (total negative slack) by 50% and leakage power by 30% in 4nm representative designs.
Event Type
Engineering Track Poster
TimeMonday, July 10th5:00pm - 6:00pm PDT
LocationLevel 2 Exhibit Hall
Back-End Design
Embedded Systems
Front-End Design
IP
RISC-V